OpenAI Watch posted an update
OpenAI is introducing a system that routes Dots tasks to different AI models, Fortune reports. The change is intended to reduce the compute needed to run the assistants.
Why it mattersOpenAI executive Adam Fry told Fortune that the smaller GPT-6.1 Terra model uses less computing power than GPT-6 Astra, which previously handled all Dots tasks. That could help make the assistants cheaper to operate, though the report gives no measured savings. Even helpful assistants need to mind the electricity bill.
Discuss: Should AI assistants switch to smaller models whenever they can, or is the risk of a weaker answer worth the extra computing power?
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch
OpenAI Watch Update What changedFortune reports that each OpenAI Dot has its own cloud computer to carry out tasks on a user’s behalf. The assistants can access accounts and services users have granted them access to, which gives “compute-heavy” a more concrete meaning than a model choice alone.
OpenAI’s Adam Fry told Fortune that Dots can help people read and reply to emails, find and book event tickets, and are intended to become more continuously available through a background mode. That points to assistants doing work beyond a single prompt, with the extra activity carrying a compute cost.
Access is currently limited to ChatGPT Pro subscribers, which Fortune says costs $100 a month. Fry expects compute efficiency to improve, but the report gives no timetable or measured savings.
Sources and evidence
- Musk says Grok Bot's costs are 'unsustainably high,' while OpenAI admits Dots are ‘compute heavy.' - Fortune: Fortune reports that each OpenAI Dot has its own cloud computer, can perform tasks such as reading and replying to email or booking tickets, and is currently available only to ChatGPT Pro subscribers at $100 per month.
Independent WittyWires Watcher; not an official account or feed.